The question comes up in every comment section now, and it never resolves.
Your partner has been talking to an AI companion. Not a search engine, not a work assistant. Something that remembers them, asks how their day went, and responds like it cares. Is that cheating?
Ask a room and you get a split. Researchers at the Université du Québec à Montréal surveyed 1,815 adults and found that roughly half considered a partner’s use of a romantic AI companion to be infidelity, and about three quarters said they would react badly to it. Which means a very large minority would not call it cheating at all, and a similar share would still be upset either way.
That is a split verdict, and a split verdict is not an answer. It is the reason the argument keeps happening.
So it is worth asking a different question, because there is one with a much clearer answer, and almost nobody is asking it.
Why the definition can never settle this
The research team defines infidelity the way most researchers do: a violation of exclusivity, sexual or emotional, that diverts romantic love, attention or investment away from a primary partner.
That is a reasonable definition, and it still cannot settle the chatbot case, because every line you try to draw runs into something on the wrong side of it.
Is it the sexual content? Then a purely affectionate companion is fine, and most people do not feel that way.
Is it the emotional investment? Then a novel that moves you is a problem, and so is a podcast host you feel close to.
Is it that there is another party? There is not one. There is a language model with no interior life, no memory of you that means anything to it, and nothing at stake.
You can keep going and you will not arrive anywhere clean. The reason is not that people are being unreasonable. It is that cheating was never a fixed category. It is an agreement between two specific people about what exclusivity means to them, and those agreements were made before this existed and do not mention it.
Which is why arguing the definition is a dead end. Nobody is going to be persuaded by a taxonomy.
The question with a clear answer
Stop asking the public whether it counts. Ask what the people doing it think.
Earlier this year the Wheatley Institute at Brigham Young University and the Institute for Family Studies published a report called Secret Soulmates, led by Brian Willoughby with Jason Carroll and Michael Toscano. They surveyed 2,431 US adults aged 18 to 30 who were dating, engaged or married.
The headline number is that 1 in 7 of those partnered young adults regularly interacts with an AI chatbot that simulates a romantic partner. Another 20% to 30% had experimented at some point. This is not a fringe behaviour and it is not going away.
But the number that answers our question is the one about disclosure.
Almost 7 in 10 regular users said it was somewhat or extremely important to them that their real life partner never learn the full extent of their use. Nearly 30% said their partner had no knowledge of it at all. Another 11% said their partner was only somewhat aware. More than half were either completely hiding it or only partly disclosing it.
The report is not called Secret Soulmates for effect.
It replicates, which is what makes it worth trusting
One survey is a finding. Two surveys, run by different teams, in different countries, with different samples, is a pattern.
Go back to the Montréal study, 1,815 Canadian adults, a completely separate research group. Nearly two thirds of regular users kept their use from their partner.
Two independent measurements of the same behaviour landed in the same place. Whatever else is uncertain here, the concealment is real.
What hiding it tells you
Here is the part worth sitting with.
People do not systematically hide things they believe are fine.
Hiding is expensive. It takes effort, it takes memory, and it carries the risk of being found out, which is worse than the original disclosure would have been. Nobody pays that cost for an activity they genuinely consider neutral. You do not clear your history after reading the news.
So when almost 7 in 10 users want their partner not to know the extent of it, they are telling you something the survey did not directly ask. They already believe their partner would object. They have run the conversation in their head, predicted the reaction, and decided against having it.
Let me be careful about the size of that claim, because there is an overstatement available and it would be false. This does not prove the behaviour is cheating. It does not tell you those partners would in fact be upset, only that the user expects it. And plenty of people keep small private things from a partner without any of it being a betrayal.
But it does dissolve the original question. You do not need a definition of cheating when the person doing it has already predicted their partner’s objection and chosen concealment over the conversation. That is not a philosophical puzzle. That is a decision somebody made.
A test that works without a definition
If you are trying to work out where your own line is, or whether something you are doing has crossed it, the definitional argument will not help you. This will.
Would you be comfortable if they read the transcript?
Not “would you show it to them.” Not “would they be entitled to see it.” Just: if they read it, would you be fine.
The value of this question is that it takes about 2 seconds to answer and you cannot argue yourself out of the result. The definitional debate is unwinnable and can be run indefinitely, which is exactly why people run it. The disclosure question refuses to cooperate with that.
And if the answer is no, the label genuinely does not matter. Something is happening that you have decided your partner cannot know about. Whether a dictionary calls it infidelity is the least useful thing about it.
What the research says actually happens
The Secret Soulmates data goes further than attitudes, and the outcomes are not subtle.
Regular use of an AI romantic companion was associated with a 46% decrease in the likelihood of being in a stable relationship, and a 40% decrease in the likelihood of having high quality communication.
Two honest caveats. This is association, not proof of cause, and it is very likely that some people turn to a companion because the relationship is already struggling rather than the other way round. Any claim that the chatbot caused the instability goes beyond what this design can show.
But one result in that data is strange enough to be worth your attention, because it is the opposite of what you would expect.
Reported relationship satisfaction was higher among users. Lower stability, worse communication, and yet people said they felt more satisfied.
That is the same shape we found in AI situationships, where the argument was that the problem with these things is not that they fail. It is that they work. A steady supply of attention that never argues, never has a bad day and never needs anything can make you feel better about a relationship while the relationship itself gets quietly worse. Satisfaction and stability came apart, and satisfaction is the one you can feel.
The finding nobody is talking about, which is more useful than the cheating question
Buried in the Montréal research is a result that reframes this whole thing.
Nearly two thirds of regular users said they wished their partner behaved more like their AI companion.
Read that again, because it is not really a story about infidelity. It is a story about an unmet want, held by a majority of users, and being expressed to a chatbot instead of to the person it is about.
That is not a rival. That is a complaint with nowhere to go.
And now put it together with the concealment. Almost 7 in 10 are hiding the use. Nearly two thirds wish their partner were different in some specific way. The hiding guarantees that the wish never gets said out loud.
So the real damage here is probably not the betrayal that everyone argues about. It is the conversation that never happens. Someone works out exactly what is missing, finds a way to get a version of it that costs nothing, and therefore never has to raise it with the one person who could actually change something.
The chatbot is not taking your partner away from you. It is absorbing the complaint that would have reached you.
One line from last week, running the other way
We argued on Monday, in going private, that couples pulling their relationship off social media are doing something healthy, and that keeping things from an audience is usually a correction rather than a loss.
This looks like the opposite advice, and it is worth being explicit about why it is not.
The direction of the withholding is the whole thing. Keeping your relationship from spectators protects it, because spectators were never entitled to it and their presence distorts it. Keeping something from your partner removes them from a decision that is partly theirs.
One reduces the audience. The other reduces the relationship to one informed person and one uninformed one. Same behaviour, opposite effect, and the test for which you are doing is simply who is being kept out.
Where this leaves VOOWA
I want to be straight about the limits here, because there is an easy claim available and it would be nonsense.
No platform can stop anyone talking to a chatbot, and any product implying otherwise is selling you something. This is not a problem an app fixes.
But it is worth naming what the companion is actually supplying, because it explains why this is hitting people in the dating stage hardest. It offers attention with the risk removed. It cannot reject you, cannot lose interest, and cannot decide after 3 months that it wants something different from what you want. Everything that makes real conversation frightening has been engineered out.
That is precisely the fear that makes early dating exhausting, and most of it is one specific uncertainty: not knowing what the other person actually wants, and not being able to ask without exposing yourself.
A platform can remove that one piece. On VOOWA every member states their marriage intention before the first message, which is the argument we made in clear coding: make the honesty structural and nobody has to be brave enough to go first. Verification handles the other half, so you are not deciding whether a person is real at the same time as deciding whether they are right.
That does not manufacture courage and it will not do the hard conversations for you. Nothing will. What it does is make sure that the highest stakes question in early dating is answered before you start, so the conversation you are avoiding is a smaller one than it used to be.
The question at the top of this piece has an answer, and it is not a definition. If you would not want them to read it, you already know. The useful move after that is not to settle the terminology. It is to say the thing you have been telling the chatbot to the person who can do something about it.
VOOWA launches on Google Play on November 11, 2026, at 11 AM. The waitlist is open below. Once you are in, invite friends with your personal code: when 11 of them join and get verified, you get a month of Basic free, and at 22, a month of Premium.
